There is a claim that people can be implicitly (unconsciously) prejudiced. Critically evaluate the evidence for this claim.
Prejudice is a generally negative attitude toward an outgroup (Brendl et al., 2001). It is an unconscious attitude and can develop either through direct experience or by social learning (Dovidio et al., 2001). It influences feelings, thoughts or actions towards social objects by automatically activating evaluations based on past experience that are not consciously remembered (Greenwald & Banaji 1995, p.8). Dovidio et al. (2001) propose four main approaches to the study of implicit prejudice: the aversive racism framework (Gaertner & Dovidio, 1986); the MODE model (Fazio, 1990); the symbolic racism framework (Sears et al., 1997) and the dual attitudes framework (Wilson et al., 2000). Various measures of implicit prejudice have been employed, including response latency, linguistic bias and physiological responses. There have been some criticisms of the existing research but, according to Dovidio et al., ‘compelling evidence has accumulated on the existence of implicit attitudes and beliefs’ (2001, p.192).
The earliest approach to the study of implicit attitudes was the aversive racism framework, proposed by Gaertner and Dovidio (1986, cited in Dovidio et al., 2001). This framework identifies three ‘cells’ of people: 1) non-prejudiced; 2) traditional racists; 3) aversive racists (Dovidio et al., 2001). It is proposed that most people wish and believe themselves to possess egalitarian attitudes but are nevertheless unconsciously negatively biased against socially disadvantaged groups.
Fazio’s MODE (Motivation and Opportunity as DEterminants of processing) model suggests that behaviour is context dependent (1990, cited in Fazio, 1995). Where there is the opportunity (e.g. time) and the motivation (e.g. social norms) to consider a response, behaviour will be influenced more by explicit than by implicit attitudes. Where there is limited or no opportunity to reflect on behaviour, or little or no motivation to deliberate, behaviour will be spontaneous and, therefore, be determined to a greater extent by implicit attitudes (Dovidio et al., 2001). In keeping with this framework, Dovidio et al. (1997) suggest that the relationship between attitudes and behaviour is dependent on the method of measurement and the type of behaviour. They identify three levels of racial attitude: public (explicit), personal (explicit) and unconscious (implicit) (Dovidio et al., 2001). Dovidio et al. (1997) propose that behaviour will be predicted by public attitudes when social desirability is a factor, personal attitudes when behaviour is private but controlled, and unconscious attitudes when the behaviour is spontaneous.
The symbolic racism framework, suggested by Sears et al. (1997) proposes that prejudice is acquired when young and retained throughout adulthood. With certain primes, prejudice can be activated and expressed in subtle ways while, explicitly, people conform to egalitarian social norms, reflecting moral codes about how they believe they should behave (Sears et al., 1997).
Wilson et al. suggest that people can hold dual attitudes, defined as ‘different evaluations of the same attitude object’ (2000, p.101). Attitudes developed in childhood become implicit when they are overridden (but not overwritten) by new attitudes that are developed later. These implicit attitudes are automatically activated and so have most influence over unconscious behaviours (e.g. body language), or behaviour that a person does not recognise needs to be controlled (Dovidio et al., 2001).
There is much evidence to support the idea that judgements can be implicitly influenced. For example, Dovidio et al. (1986, cited in Dovidio et al., 2001) found evidence for automatic stereotyping using a response-latency task. Participants were presented with either the word ‘white’ or ‘black’ and then given a second word judged to be either a positive or negative stereotypical attribute of the black and white ethnic groups. Participants were asked if the attribute could ever be true or was never true of the group. Reaction times were found to be faster for attributes judged to be stereotypical of that group.
Devine (1989) presented half of her participants with 80 words stereotypically associated with black people and 20 unconnected words. The remaining participants were presented with the reverse. Participants were then shown a fictitious character and asked to rate its various actions. It was found that the participants who had been shown more words stereotyping black people rated the character as being more aggressive. Devine (1989) suggested that activating the stereotype caused participants to project stereotypical attributes of black people onto the character. However, Devine’s (1989) findings have been criticised. For example, many of the participants may have imagined the fictitious character to be white and, therefore, it is possible that the words primed hostility directly, rather than a stereotype of black people, as intended (Greenwald & Banaji, 1995). Many of the words presumed to be stereotypical of black people were generally negative and could have activated aggression directly (Greenwald & Banaji, 1995; Hamilton & Sherman, 1994 cited in Dovidio et al., 2001 and Lepore & Brown, 1997).
Furthermore, the presence of stereotyping does not necessarily indicate prejudice. For example, Devine (1989) distinguished between knowledge of a stereotype and endorsement of a stereotype, with the level of endorsement indicating prejudice. However, in the experiment by Dovidio et al. (1986, cited in Dovidio et al., 2001), it was observed that reaction times were faster for positive attributes paired with ‘white’ than those paired with ‘black’, and for negative attributes paired with ‘black’ than those paired with ‘white’. This finding is confirmed by the study of Greenwald et al. (1998: Experiment 3), where participants were asked to classify black versus white names and pleasant versus unpleasant words. They found that participants responded faster for white-positive pairings than for black-positive pairings. Similarly, negative implicit attitudes of white people towards black people were observed in studies by Ottaway et al. (2001, cited in Dovidio et al., 2001) and Rudman et al., (1999 cited in Dovidio et al., 2001).
Perdue et al. (1990, cited in Dovidio et al., 2001) found that there was automatic bias for one’s ingroup. In this study, participants were subliminally primed with pronouns (e.g. we, us, them) and then presented with a target word that they had to judge as being positive or negative. Reaction times were faster for positive words after ingroup priming (e.g. we, us) than outgroup priming (e.g. them), indicating that participants associated positive words more strongly with their ingroup than with outgroups. Maass et al. (1989, cited in Dovidio et al., 2001) measured linguistic intergroup bias and found that undesirable actions by outgroup members were encoded at a more abstract level than those by ingroup members; i.e., a general judgement was made about outgroup members’ characters based on the actions in question. However, desirable actions were encoded at a more concrete level for outgroup relative to ingroup members: i.e., the behaviour was held to be situation specific. This allows inconsistent stereotyping behaviours by outgroup members to be disregarded, protecting more general beliefs held about the outgroup (Dovidio et al., 2001).
Lepore and Brown (1997) state that it is important to distinguish between categories (e.g. ethnic-group labels) and stereotypes (e.g. character traits) when using primes. The level of participants’ prejudice will not affect results when traits are used as primes, as these traits have the effect of priming negative stereotype knowledge, irrespective of attitude. Wittenbrink et al. (1997) found that the level of prejudice does have an effect when ethnic groups are used as primes. They found that negative words were more highly associated with a black person and positive words with a white person and that the effect correlated with the level of participants’ prejudice, as measured by explicit racial-attitude scores. Non-prejudiced people cannot control the automatic activation of stereotyped knowledge, but they will consciously suppress it.
Brendl et al. (2001) found that non-words were judged in a similar way to negative items and suggest that longer response times may be the result of non-familiarity. They reflect the relative ease of retrieving pre-stored evaluations for familiar versus less familiar stimuli. Dasgupta et al. (2000) tested this hypothesis with reference to the relative familiarity of white stereotypes in a white-dominant society (e.g. America). Their findings showed that positive attributes were still more strongly associated with white than black Americans when statistically controlling for differences in the familiarity of stimuli. Ashburn-Nardo et al. (2001) found ingroup bias in their experiments, even though participants had no experience of the ingroup and/or outgroup. They suggest that this bias may reflect ingroup favouritism rather than prejudice against outgroups.
Many studies have investigated the relationship between implicit and explicit prejudice, but there is no clear understanding of what the relationship is (Blair, 2001, cited in Dovidio et al., 2001). Banaji and Greenwald (1995) investigated gender bias in judgements of fame by asking participants to rate the fame of male and female names. They found a bias toward male names, with a lower criteria required in judging familiar male names than required in judging female names.
A meta-analysis by Dovidio et al. (2001) reviewed 27 studies that looked at racial attitudes and found that there was a significant, but weak, correlation between implicit and explicit attitudes across 14 tests using priming measures, three tests using other latency measures (e.g. time spent looking at pictures) and four tests using physiological measures. The correlation was stronger for topics that were not socially sensitive than for those that were. For example, Fazio et al. (1990, cited in Dovidio et al., 2001) found a high correlation for topics such as dentists, but a weak correlation for topics such as pornography. Stacy (1997, cited in Dovidio et al., 2001) found that implicit attitudes predicted marijuana use while explicit attitudes did not, while they both predicted alcohol use, which is less socially stigmatised (Dovidio et al., 2001).
Dovidio et al. found that implicit measures predicted spontaneous but not deliberate behaviours (1997: Experiment 2). They asked participants to interact face-to-face with both black and white partners and to then rate both partners on rating scales. Participants’ rates of blinking and their eye contact with the partners were also measured. It was found that deliberate behaviours (i.e. the ratings) were correlated with explicit measures of self-reported prejudice but not with implicit measures. However, implicit measures predicted participants’ non-verbal behaviour (i.e. blinking and eye contact); negative attitudes predicted more blinking and less eye contact (1997: Experiment 3). Furthermore, a study by Fazio et al. (1995) looking into the Rodney King verdict and perceptions of the black community’s resultant anger showed that explicit measures (i.e. direct ratings of the verdict and the reaction) were correlated with self-reported prejudice. The implicit measures did not correlate with these ratings; however they correlated more highly than the explicit measures when participants were asked to rate the relative responsibilities of the white and black communities for the post-verdict reaction. Crosby et al. (1980, cited in Dovidio et al., 1997) suggest that unconscious, nonverbal behaviours may be less subject to social desirability than verbal behaviours. Fazio et al. (1995) suggested that non-verbal behaviours are liable to ‘leakage’; i.e. they occur in spite of conscious efforts to appear non-prejudiced.
One proposed explanation for the weak and variable relationships between implicit and explicit measures is poor reliability of the implicit measures (Kawakami & Dovidio, 2001). However there are many other theories. For example, it may have more to do with the nature of what is being measured (Dovidio et al., 2001). Specifically implicit measures usually require global evaluations (i.e. positive versus negative), while explicit measures can be more complex (e.g. McConahay’s Modern Racism Scale, 1986, cited in Dovidio et al., 2001). Dovidio et al.’s meta-analysis (2001) found a higher relationship between implicit and explicit prejudice for the nine tests requiring general evaluations than for the 18 using more issue-oriented measures.
There is also a theory that implicit and explicit attitudes use different processes and, therefore, would not be expected to show a high correlation (Dovidio et al., 2001; Karpinski & Hilton, 2001). The automatic activation of an evaluation does not necessarily mean that the evaluation will be used (Fiske, 1989 and Gilbert & Hixton 1991, both cited in Dovidio et al., 2001). It has been suggested that response latency tasks measure the time taken to activate evaluations, while self-report tasks measure the use of these evaluations in making judgements (Dovidio et al., 2001). Devine (1989) said that low-prejudiced people will attempt to control the use of these biased evaluations and even prejudiced people may be motivated to conform to egalitarian social norms (Dovidio et al., 2001). For example, Fazio et al. (1990, cited in Dovidio et al., 200) found a weak negative correlation between implicit and explicit attitudes for socially sensitive objects and a high positive correlation for socially non-sensitive objects. Fazio et al. (1995) found a higher correlation between implicit and explicit measures among participants who were less motivated to control their prejudice.
Research to increase the understanding of the psychometric properties of implicit measures may help to account for the variable correlations observed between implicit and explicit attitudes (Fazio et al., 1995). A common measurement technique is the use of response latency tasks combined with priming. In these tasks a social category is primed, often subliminally, and then a target word is presented for participants to judge as either positive or negative. A faster response would demonstrate a stronger association between the two concepts being presented (Dovidio et al., 2001; Karpinski & Hilton, 2001). Specifically, it is expected that participants will respond faster to positive words when their ingroup has been primed than when their outgroup has been primed, and vice versa. The IAT (Implicit Association Test, Greenwald et al. (1998) is a frequently used response latency measure and the ‘best developed measure of implicit evaluations’ (Brendl et al., 2001 p.7600). This test requires participants to judge data presented on the computer’s screen, by classifying it according to evaluative categories (e.g., positive or negative), and social categories (e.g., black or white). Participants use the computer keys to record their responses. It is expected that prejudiced participants will respond faster when, for example, ‘white’ and ‘positive’ share a response key (an ‘evaluatively compatible’ condition) than when ‘white’ and ‘negative’ share a response key (an ‘evaluatively incompatible’ condition). Longer response times indicate greater implicit prejudice as they imply a weaker association between the two concepts (Greenwald et al., 1998; Dovidio et al., 2001). Gawronski, et al. (2007) reviewed three key assumptions about implicit attitudes: that they 1) reflect unconscious representations; 2) are less susceptible to social desirability than self-reports; 3) reflect older and more stable representations originating in socialisation. Their review found no clear support for these assumptions and led them to propose an alternative model. This model suggests that indirect measures reflect the activation of associations from memory, and that these associations can occur irrespective of whether or not a person considers them to be accurate. In addition, Brendl et al. (2001) state that the IAT may simply measure relative attitudes towards two groups and therefore, it is possible that although one group is preferred to the other, neither group is actually negatively evaluated.
Various studies using the IAT have found negative implicit attitudes held by white people toward black people (e.g. Dasgupta et al., 2000; Greenwald et al., Experiment 3, 1998; Ottaway, Hayden & Oakes, 2001 cited in Dasgupta et al., 2000; Rudman et al., 1999 cited in Dovidio et al., 2001). However, the reliability and validity of the IAT has been called into question, e.g. by Dovidio et al. (2001), who found in their meta-analysis that the test-retest reliability of the IAT was only moderate. Other studies into implicit stereotypes have shown stability over time. For example, Kawakami and Dovidio (2001) found consistently faster response times in their experiments for stereotype-consistent than stereotype-inconsistent trait photographs over periods ranging from one hour to 15 days. Rudman et al. (1999, cited in Dovidio et al., 2001) found reliable racial stereotyping over a nine-week period. Test-retest reliability has also been achieved in studies on racial and gender stereotypes by Kawakami and Dovidio (2001) and on self-regard by Pelham and Hetts (1999, cited in Dovidio et al., 2001).
Dovidio et al. (2001) question the internal consistency of the IAT. For example, Dasgupta et al. (2000) found low correlation between implicit racial attitudes when stereotyped white or black names were used and when photographs of white or black people were used (cited in Dovidio et al., 2001). Further, Sherman et al. (1999, cited in Dovidio et al., 2001) reported significantly different correlations between the IAT and a priming measure in two different studies (Dovidio et al., 2001). However, De Houver (1999, cited in Dovidio et al., 2001) states that different measures have fundamentally different structures and call upon different cognitive structures. Therefore, a high correlation between different measures would not be expected (Greenwald et al., 1998; Dovidio et al., 2001) and would, in fact, indicate poor discriminant validity (Campbell & Fiske, 1959; Greenwald et al., 1998).
Reliability and validity are vital if we are to assume that implicit measures represent real attitudes (Robinson et al., 1991 cited in Dovidio et al., 2001). Cameron et al. (2000, cited in Karpinski & Hilton, 2001) were unable to correlate the IAT with various other implicit measures, suggesting that the IAT may, in fact, not be measuring attitudes at all, but merely associations that an individual has been previously exposed to (Karpinski & Hilton, 2001) Therefore, a tendency to associate white more with positive and black more with negative may indicate higher exposure to these associations in the form of cultural stereotypes, rather than higher levels of prejudice (Greenwald et al., 1998: Experiments 2-3; Karpinski & Hilton, 2001). This follows on from the point made by Devine (1989), mentioned earlier, when she distinguished between knowledge of and endorsement of stereotypes. Fiedler, Messner and Bluemke (2006, cited in Dovidio et al., 2001) agree that it is incorrect to assume that recognition of a close association between two concepts equates to an attitude. In addition, they point out that the term ‘black’ is used often as a negative term where race is not involved (e.g. black sheep), and so the word itself may hold negative connotations that have nothing to do with attitudes toward black people. It is necessary to have confidence in the measures employed before they can be used to test relationships (Campbell & Fiske, 1959). Implicit measures require psychometric testing to identify their meaning (Campbell & Fiske, 1959 cited in Dovidio et al., 2001).
In summary, the reliability and validity of the IAT has been found to be moderate. It may be necessary to improve understanding of implicit processes and their relationships with different measurement techniques (Dovidio et al., 2001). This improved understanding of the psychometric properties of implicit measures is necessary if they are to be used for predicting behavior (Fazio et al., 1995).
Further research into the psychometric properties of implicit measures may be necessary; it will be important to determine whether implicit measures always succeed in measuring attitudes, rather than simply associations. Implicit and explicit measures of prejudice appear to have only a weak relationship, but this relationship is not fully understood. Generally, research supports the existence of implicit prejudice and indicates that it can predict behaviour, especially that which is spontaneous and independent of explicit attitudes.
Ashburn-Nardo, L., Voils, C.I. and Monteith, M.J. (2001) Implicit associations as the seeds of intergroup bias: how easily do they take root? Journal of Personality and Social Psychology, Vol. 81(5), 789-799
Banaji, M.R. and Greenwald, A.G. (1995) Implicit Gender Stereotyping in Judgments of fame. Journal of Personality and Social Psychology, Vol. 68 (2), 181-198
Brendl, C.M., Markman, A.B. and Messner, C. (2001) How do indirect measures of evaluation work? Evaluating the inference of prejudice in the Implicit Association Test. Journal of Personality and Social Psychology, Vol. 81(5), 760-773
Campbell, D.T. and Fiske, D.W. (1959) Convergent and discriminant validation by the multitrait-multimethod matrix. Psychological Bulletin Vol. 56(2), 81-105
Dasgupta, N., McGhee, D.E., Greenwald, A.G. and Banaji, M.R. (2000) Automatic Preference for White Americans: Eliminating the Familiarity Explanation. Journal of Experimental Social Psychology, Vol. 36, 316-328
Devine, P.G. (1989) Stereotypes and Prejudice: Their Automatic and Controlled Components. Journal of Personality and Social Psychology, Vol. 56 (1), 5-18
Devine, P.G. (2001) Implicit prejudice and stereotyping: how automatic are they? Introduction to the special section. Journal of Personality and Social Psychology, Vol. 81(5), 757-759
Dovidio, J.F., Kawakami, K., Johnson, C., Johnson, B. and Howard, A. (1997) On the Nature of Prejudice: Automatic and Controlled Processes. Journal of Experimental Social Psychology, Vol.33, 510-540.
Dovidio, J.F., Kawakami, K. and Beach, K.R. (2001) Implicit and Explicit Attitudes: Examination of the relationship between measures of intergroup bias. In R. Brown and S. Gaertner (Eds.), Blackwell Handbook of Social Psychology: Intergroup Processes. Oxford: Blackwell.
Fazio, R.H., Jackson, J.R., Dunton, B.C. and Williams, C.J. (1995) Variability in Automatic Activation as an Unobtrusive Measure of Racial Attitudes: A Bona Fide Pipeline? Journal of Personality and Social Psychology, Vol. 69(6), 1013-1027
Gawronski, B., LeBel, E.P. and Peters, K.R. (2007) What Do Implicit Measures Tell Us? Perspectives on Psychological Science, Vol. 2(2), 181-193
Greenwald, A.G. and Banaji, M.R. (1995) Implicit Social Cognition: Attitudes, Self-Esteem, and Stereotypes. Psychological Review, Vol. 102(1), 4-27
Greenwald, A.G., McGhee, D.E. and Schwartz, J.L.K. (1998) Measuring Individual Differences in Implicit Cognition: The Implicit Association Test. Journal of Personality and Social Psychology, Vol. 74(6), 1464-1480
Karpinski, A. and Hilton, J. (2001) Attitudes and the Implicit Association Test. Journal of Personality and Social Psychology, Vol. 81(5), 774-788
Kawakami, K. and Dovidio, J.F. (2001) The Reliability of Implicit Stereotyping. Personality and Social Psychology Bulletin, Vol. 27, 212-225
Lepore, L. and Brown, R. (1997) Category and Stereotype Activation: Is Prejudice Inevitable? Journal of Personality and Social Psychology, Vol. 72(2), 275-287
Sears, D.O., Van Laar, C., Carrillo, M and Kosterman, R. (1997) Is It Really Racism? The Origins of White Americans’ Opposition to Race-Targeted Policies. The Public Opinion Quarterly, Vol. 61(1), Special Issue on Race, 16-53
Wilson, T.D., Lindsey, S. and Schooler, T.Y. (2000) A Model of Dual Attitudes. Psychological Review, Vol. 107(1), 101-126
Wittenbrink, B. Judd, C.M. and Park, B. (1997) Spontaneous Prejudice in Context: Variability in Automatically Activated Attitudes. Journal of Personality and Social Psychology, Vol. 81(5), 815-827